Developing computer systems that can recognize arguments in natural human language and engage in debate is a major challenge. One of the most demanding challenges in the field of artificial intelligence (AI).
On Nature now scientist Noam Slonim, Israeli researcher at IBM, reports an impressive development in this field: Project Debater, an artificial intelligence system that can engage with humans in the debate. An AI that knows how to dispute.
The results, surprising in their own way, show how far research has come in this area.
Less than a decade ago, analyzing human speech to identify the ways in which a theme is used to support conclusions was well beyond computational capabilities.
Since then, a combination of technical advances in artificial intelligence and increasing maturity in engineering (alongside growing commercial demand) has led to rapid expansion.
More than 50 labs around the world are working on the problem, including teams from all major software companies.
Because it is not at all easy to develop an artificial intelligence capable of having a debate with humans
The structure of a controversy, of an argument, is too varied, too complex, too nuanced and often too veiled to be easily recognized. For this Slonim has decided to start a great challenge: to develop a completely autonomous system that can take part in a live debate with humans.
Project Debater is the culmination of this work.
Project Debater is, first of all, an extraordinary engineering feat.
An extraordinarily ambitious undertaking. And as with almost all AI research that aims this high, a key bottleneck is data acquisition. You need enough of them to be able to calculate an effective solution to the challenge of making a machine can argue.
Project Debater has faced this obstacle. It did so using a two-pronged approach. Primo, he narrowed his attention to about 100 topics of debate. Secondo, collects its raw material from large data sets, even by the standards of modern language processing systems.
In a series of releases in 2018 and 2019, Project Debater tackled a number of talented and high-profile human debates, and her performance was informally evaluated by the audience.
What is the outline of a debate with Project Debater?
Supported by its argumentation techniques and powered by its processed datasets, the system creates a 4-minute speech that opens a debate on a topic from its repertoire, to which a human opponent responds. He then reacts to his opponent's points by producing a second 4-minute speech. The opponent responds with their own 4-minute rebuttal, and the debate ends with both participants giving a 2-minute closing statement.
Perhaps the weakest aspect of the system is that it struggles to emulate the coherence and flow of human debates.
Yet this limitation is certainly not Project Debater's "fault".
The structure of the topic is still little known, despite two millennia of research. Depending on whether the focus of the research is language use, epistemology (the philosophical theory of knowledge), cognitive processes, or logical validity, the features proposed as crucial to a coherent model of argumentation and reasoning differ markedly.
The models of what constitutes a good argument are therefore extremely diverse. By contrast, models of what constitutes good debate amount to little more than formalized insights (although disciplines in which the goodness of debate is codified, like law, have an advantage on this front).
It is therefore no wonder that Project Debater's performance was evaluated simply by asking a human audience whether they thought it was “an example of a decent debate.” For nearly two-thirds of the topics discussed, humans thought this was the case.
The Debater project is a crucial step in the development of topic technology. These findings offer a tantalizing look at how an AI system might work. With what? With the network of arguments that human beings interpret with so much apparent ease.
Between fake news and polarization of public opinion, the need for human beings to be supported in the creation, processing and sharing of complex topics could soon become pressing.
AI could "train" human beings in debate (and critical thinking).
In summary: Project Debater's challenge is hard, but it also represents progress that can contribute to human reasoning.